OpenAI keeps working to improve the user experience by making processes better and launching updates that bring more refinement. Such was the intention when the company deployed an update to ChatGPT, but the update, instead of creating the intended experience, ended up providing strange responses to prompts that were not coherent and appeared confusing and loopy. Models acting up after an update is not an uncommon scenario, as a bug can often appear. However, the issue became more widespread, and the responses became more strange, leading the company to call for a rewind on the feature.
OpenAI's update to the GPT-4o model led to some strange responses, leading the company to dial back on it
OpenAI released a GPT-4o update last week that was meant to enhance the model by making it more intelligent and adding more personalization, but what was meant to give it a personality ended up stepping in the direction of overpersonalization as the model ended up being too sycophantic and overly flattering or polite which annoyed many of the users.
The users then took to Reddit and other forums to express their frustration over the update that made the assistant less authentic instead of offering a human element. It did not take long for Altman to acknowledge the issue and address that the chatbot's personality had shifted in an unsettling way. Altman, while acknowledging the issue, said:
The last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week. at some point will share our learnings from this, it’s been interesting.
This is just one example of how unpredictable AI can be and how optimization can sometimes take away the user experience you have been chasing. After the update, ChatGPT started responding to every prompt in an inauthentic manner that involved robotic praises and the tone felt too forced.
OpenAI while explaining the root cause of the problem, pointed out to a system that relied more on short term user feedback loops rather long-term satisfaction leading to responses that were not genuine and model being more flattering and less informative. OpenAI is said to have rolled back on the update completely and bringing the GPT-4o back to having a more balanced tone.
OpenAI aimed to make ChatGPT more supportive but adding an authentic personality to the model is a complex task and looking ahead, OpenAI aims to offer default personality settings and ability to customize personalities based on user preference.
Follow Wccftech on Google to get more of our news coverage in your feeds.
